Source sector and region contributions to BC and PM 2.5 in Central Asia and

Particulate matter (PM) mass concentrations, seasonal cycles, source and source region contributions in Central Asia (CA) are analyzed for the period April 2008– July 2009 using the Sulfur Transport and dEposition Model (STEM) chemical transport model and modeled meteorology from the Weather Research and Forecasting (WRF) 5 model. Predicted Aerosol Optical Depth (AOD) values (annual mean value ∼ 0.2) in CA vary seasonally with lowest values in the winter. Surface PM 2.5 concentrations (annual mean value ∼ 10 µgm − 3 ) also exhibit a seasonal cycle, with peak values and largest variability in the spring/summer, and lowest values and variability in the winter (hourly values from 2–90 µgm − 3 ). Surface concentrations of black carbon (BC) (mean value 10 ∼ 0.1 µgm − 3 ) show peak values in the winter. The simulated values are compared to surface measurements of AOD, and PM 2.5 , PM 10 , BC, organic carbon (OC) mass concentrations at two regional sites in the Kyrgyz Republic (Lidar Station Teplokluchenka (LST) and Bishkek). The predicted values of AOD and PM mass concentrations and their seasonal cycles are fairly well captured. The carbonaceous aerosols are under- 15 predicted in winter, and analysis suggests that the winter heating emissions are underestimated in the current inventory. most important to the anthropogenic portion of combustion and are shown to be the most for BC. the region also PM and BC The analysis of the transport pathways the variations in particulate and in CA demonstrate that this region is strategically located to characterize regional and intercontinental trans- port of pollutants. Aerosols at these sites are shown to reﬂect dust, biomass burning and anthropogenic sources from Europe, South, East and CA, and Russia depending on the time period. from wind-blown dust, open biomass burning, and anthropogenic sources, and dif-ferent geographical source regions and source sectors (transportation, power, industry and residential). The simulated values are compared to surface measurements of AOD, 20 PM 2.5 , PM 10 , BC, OC mass concentrations at the two regional sites in CA. The transport of aerosols into CA is also explored through three dimensional backward trajectory analysis. Transport from CA and their impacts on downwind areas are also analyzed via forward trajectory analysis. Finally we present results of how the PM concentrations may change in the future using emission scenarios for 2030 that reﬂect possible 25 air quality and climate policies.

indicated that the major transport pathway of pollution from Europe to Asia is via low altitude flows passing through CA. The magnitude of the pollution transport from Europe to Asia is highly uncertain in large part due to the lack of observations of pollutants along this pathway. To help better characterize the air pollution levels and the transport pathways in the region a study was undertaken between Russia, Kyrgyz Republic, and 5 USA scientists to observe and model aerosols in the region. Measurements of particulate matter (PM) mass and composition were taken at two locations in the Kyrgyz Republic (Lidar Station Teplokluchenka (LST) and Bishkek) and modeling analysis was performed to assess the contributions of local, regional and distant sources to the PM concentrations in the region (Miller-Schulze et al., 2012;Chen et al., 2012Chen et al., , 2013. 10 In this paper we present a modeling analysis of PM 2.5 , PM 10 , (PM 2.5 refers to particles in the size range of less than 2.5 µm aerodynamic diameter (AD) and PM 10 refers to particles in the size range of less than 10 µm AD), black carbon (BC) and organic carbon (OC) mass concentrations and aerosol optical depth (AOD) over the time period of April 2008 to July 2009. The Sulfur Transport and dEpostion Model (STEM), 15 a hemispheric chemical transport model (D'Allura et al., 2011), is used to estimate spatial and temporal variations in PM in CA, and to assess the contributions to PM from wind-blown dust, open biomass burning, and anthropogenic sources, and different geographical source regions and source sectors (transportation, power, industry and residential). The simulated values are compared to surface measurements of AOD, 20 PM 2.5 , PM 10 , BC, OC mass concentrations at the two regional sites in CA. The transport of aerosols into CA is also explored through three dimensional backward trajectory analysis. Transport from CA and their impacts on downwind areas are also analyzed via forward trajectory analysis. Finally we present results of how the PM concentrations may change in the future using emission scenarios for 2030 that reflect possible 25 air quality and climate policies.
11347 LST (42 • 27 49.38 N, 78 • 31 44.17 E, elevation 1921 m a.s.l.) sites are denoted by circle and triangle markers, respectively, in Fig. 1. Both sampling sites are in mountain ranges with valleys to the north, with mountains that reach elevations greater than 3500 m a.s.l. south of the Bishkek site and 4600 m a.s.l. south of the LST site, and essentially no population to the south. At each site, PM 2.5 mass was measured contin- 10 uously with tapered element oscillating microbalance (TEOM) instruments and PM 2.5 , PM 10 , BC, and OC were obtained using filter-based sampling with samples collected for 24 h every other day. AOD was measured every day at 10.30 a.m. local time (LT) using Microtops-II sun-photometers (SP). A stationary three wavelength aerosol Lidar measured vertical profiles of extinction and depolarization on an event basis at the 15 LST site. The Lidar vertical profiles provide information on vertical distribution of the particles, and were also used to calculate AOD from the Lidar Extinction (LE) profiles and to estimate the height of the planetary boundary layer (PBL) as described in Chen et al. (2013). These observations sites are now part of the UNEP project ABC measurement network (http://www.rrcap.ait.asia/abc/index.cfm). Further details of the study can 20 be found in (Miller-Schulze et al., 2011). Observations from these sites were obtained for the period April 2008 to July 2009 (the TEOM measurements were available from April 2008 and filter measurements began from 1 July 2008).
The Moderate Resolution Imaging Spectroradiometer (MODIS) collection 5.1 Level 2 AOD products (∼ 10 km horizontal resolution) at 550 nm wavelength from Terra and 25 Aqua satellites were used to compare the observed and simulated AOD. The MODIS Level 2 data were used and included land and ocean AOD retrieved via the dark target algorithm (Remer et al., 2005;Levy et al., 2007), and the Deep Blue AOD over land 11348 Nitrate and secondary organic aerosols (SOA) were not included in the model for this application. The nitrate aerosol is estimated to be a minor component of the PM mass in CA (Baurer et al., 2007). The importance of SOA will be discussed later in the paper. The dry deposition of aerosols was modeled using the "Resistance in Series Parameterization" (Wesely and Hicks, 2000) and wet deposition was calculated as a loss rate 5 based on the hourly precipitation calculated from the WRF model. Further details of the wet scavenging can be found in Adhikary et al. (2007). The modeled AOD at 550 nm wavelength was calculated using the simulated three dimensional aerosol distributions and species specific extinction coefficients as described in Chung et al. (2010). 10 The STEM and WRF computation domains were identical, with a 60 km×60 km horizontal resolution (249 × 249 horizontal grid cells) and 22 vertical layers up to 10 hPa. The domain ( Fig. 1) covered much of the Northern Hemisphere in a polar stereographic projection, centered over the Arctic region and extended to 35 • N to include the major emission regions of North America, Europe, and Asia. This modeling system has 15 been applied to simulate aerosol distributions in several field campaigns as described in D' Allura et al. (2011) and further details describing the model can be found there. STEM was initialized with a one month spin up using March 2008. Much of the analysis for this paper is focused on the domain denoted by the rectangle centered over CA shown in Fig. 1. This domain has large gradients in topography (insert Fig. 1), which 20 significantly impact the transport patterns in the region.

Air mass trajectories
The CA observation sites are impacted by dust, anthropogenic pollution, and biomass burning emissions from various source regions. To further understand the transport pathways and source region influences on the PM distributions at these sites, three di-25 mensional ten day air mass trajectories (both forward and backward in time) from each  April 2008-July 2009. In this trajectory analysis, we utilized the three dimensional wind fields (including u, v and w components) along with the above ground level (a.g.l.) altitude simulated by the WRF meteorological model consistent with Dallura et al., 2011 study. These trajectories describe the general flow patterns based on wind fields alone and provide useful information 5 about the history of air mass particularly the influence of source regions over which the air mass had resided before arriving at the site of interest. Note that these trajectories do not account for any other atmospheric processes such as diffusion or chemical evolution along its path (Kurata et al., 2004 andGuttikunda et al., 2005).
To understand the differences in transport patterns at the surface and aloft, and 10 to study the impact of topographic gradients in the vicinity of the sites, trajectories were initialized at different altitudes (0.1 (100 m), 0.3 (300 m) 0.5 (500 m), 1, 2, 3, and 5 km a.g.l.) at the site locations (i.e. latitude and longitude) daily every 3 h for a ten day period both backward and forward in time. The trajectories were terminated when they touched the ground, or went out of the model domain or exceeded the ten day 15 calculation period. The trajectories (at or below 1 km) were used to characterize transport pathways impacting the surface concentrations at these sites, which are discussed later in Sect. 3.5.  Visschedijk et al., 2009;Denier van der Gon et al., 2009). The shipping emissions came from the IIASA base year 2005 inventory (UNEP and WMO, 2011). Mass conservative regrinding tools including MTXCALC and MTXCPLE from the IOAPI m3tools suite (http://www.baronams.com/products/ioapi/AA.html#tools) were used to interpolate the input raw emissions described above on to the model grid.

Emissions
Anthropogenic emissions for SO 2 , BC and OC were available by major economic sectors; i.e., transportation, residential, industry, and power. The industry and power sectors were treated as small and large point sources, respectively, and emitted into the first 6 model levels (lowest 2 km). The residential and transportation emissions were treated as area sources and partitioned into the first two model levels with a 90-10 % 10 split. Monthly emission allocation factors were applied over India and China for the economic sectors from Lu et al. (2011). The rest of the domain (i.e. excluding India and China) used same emission rates for all months due to unavailability of monthly emission allocation factors.
The Fire Inventory from NCAR (FINN v1) was used for BC, OC, CO, SO 2 , PM 2.5 15 and PM 10 biomass burning emissions from forest, grassland and crop residual fires. The FINN database, which is based on MODIS fire detection as thermal anomalies, provides global coverage of fire emissions at a spatial resolution of ∼ 1 km on a daily timescale (Wiedinmyer et al., 2011 WRF output) and only grid cells with snow cover < 1 % were used for dust emission calculations. Figure 2 shows the annual gridded anthropogenic SO 2 and BC, dust, and biomass burning PM 2.5 emissions in Gg per grid in and around CA. Large BC emission hotspots can be seen over the Indo-Gangetic plain and eastern China. Significant BC emissions 5 are also seen over Europe, but are relatively lower in intensity than the Asian sources. The SO 2 emissions show Eastern China as the largest source region followed by regions of South Asia, Europe, and Russia. The major natural dust emission sources ( Fig. 2c) include Africa, the Middle East, CA, Western India boundaries, and Western China. The major sources of biomass burning are Eastern Europe, portions of Siberian The open biomass burning emissions that impact CA also have a strong seasonality with minimum impact in winter (Supplement Fig. S1). Fires typically begin in the spring 20 in Siberia along 50 • N latitude and in northern India and South East Asia and in summer the high latitude burning shifts to the west. In October the fire activity decreases and remains low until spring, with the most active fire regions associated with agricultural burning in northern India and southeast China.

25
In addition to the base emissions, a series of simulations were analyzed using emission scenarios for 2030. These scenarios were developed for the WMO/UNEP report that looked at short lived climate pollutants as described in Shindell et al. (2012) Anenberg et al. (2012). The reference scenario for 2030 was based on the implementation of control measures currently approved in the various regions and assumed their perfect implementation. The 2030 reference scenarios were developed from a reference global emissions inventory with a 2005 reference year, and assumed significant growth in fossil fuel use relative to 2005, leading to increases in estimated CO 2 emis-5 sions (45 %). Abatement measures prescribed in current legislation were projected to lead to reductions in air pollutant emissions, which varied by pollutant and region.
In the 2030 reference scenario, total primary PM 2.5 emissions remain approximately constant, while BC and OC decline by a few percent. However, in the study domain emission changes varied widely. BC emissions increased by 10-100 % in CA, South 10 and Southeast Asia and in western China, and decreased in East Asia and Europe. The PM 2.5 emissions showed similar regional changes but grew at smaller rates (10-40 %). SO 2 emissions generally increased throughout the region by 10-20 %. Spatial maps of emission changes for the 2030 reference scenario are presented in Figs. S2b, S3b, and S4b.

15
A series of emission control scenarios for 2030 were developed to evaluate the impact of additional abatement measures designed to reduce the levels of short lived climate pollutants (e.g., BC). The BC measures in the scenarios included two different sets of assumptions (low and lowest). The first focused on reductions from incomplete combustion sources. These included implementation of Euro 6 equivalent vehi-20 cle emission standards (requiring installation of diesel particulate filters) and improving traditional biomass cook stoves in developing countries (assuming 25 % decrease in BC and 80-90 % decreases in OC, CO, non-methane volatile organic compounds (NMVOC), methane, and direct PM 2.5 , relative to emissions from traditional stoves). Under this scenario BC and PM 2.5 emissions in the study region are projected to de- The lowest option assumed the additional elimination of high-emitting vehicles, biomass cook stoves (in developing countries), and agricultural waste burning. These Introduction These measures were also combined with a scenario designed to stabilize greenhouse gases at 450 ppm of CO 2 equivalent (lowest + 450 ppm scenario), consistent with a global average temperature increase of ∼ 2 • C. These CO 2 measures reduced SO 2 (−30 %) (Fig. S4d) and NO x (−20 %), but had little further impact on BC (∼ 5 % decline, Fig. S2d) since the major sources of CO 2 differ from those of BC. PM 2.5 emissions were substantially further reduced under this scenario (Fig. S3d).

Simulations analyzed
Several simulations were analyzed for this paper. The base simulation included all sources and used the meteorology from the WRF model for the period April 2008-July 2009. To investigate the contributions from specific source sectors, additional sim- 15 ulations were performed where emissions from one sector were set to zero everywhere. The contribution from each sector was calculated as the difference between the base simulation and the simulation with emissions from that particular sector set to zero. This was repeated for each sector and for biomass burning. Additional simulations were performed to assess the source contribution from specific regions to the particle 20 levels in CA. The specific regions used are shown in Fig. 1. In these simulations all anthropogenic emissions were set to zero in that region. In a similar manner regional dust and fire sources were also studied and the source regions are also shown in Fig. 1

Regional perspective
CA is a region with high aerosol loadings as shown in the mean MODIS retrieved AOD at 550 nm for the time period of study (April 2008-July 2009) (Fig. 3). AOD (period mean) throughout CA (∼ 45-90 • E, 35-50 • N) are greater than 0.25, with the highest 5 regional values around the desert areas near the Caspian and the Aral seas. There are also high values (> 0.6) along CA's eastern border, which reflect the deserts and rapidly developing cities in western China, and to the south over Pakistan and northern India.
The predicted period mean AOD spatial distribution shown in Fig. 4d captures the 10 main observed features. The period-mean predicted surface concentrations of PM 2.5 , BC, and total dust (fine and coarse) are also plotted. The period mean PM 2.5 concentrations in CA (10 to 35 µg m −3 ) have a similar geographical distribution as AOD.
Dust is the major component of predicted PM 10 in CA and concentrations are high (25-100 µg m −3 ). The BC levels in CA are typically less than 0.3 µg m −3 and its spatial 15 pattern reflects contributions from both anthropogenic and biomass burning sources.

Comparison with surface observations in CA
The surface observations at the two CA sites provide the opportunity for the first time to evaluate the performance of chemical transport models in estimating the distribution of aerosols in CA and to assess the emission estimates in the region. A comparison of 20 the predicted and observed meteorology is presented in Fig. 5, where the distributions of key meteorological parameters for the entire measurement period are shown as box-plots. The model accurately predicted the magnitude and variability in temperature and relative humidity. For example the model mean value of temperature and relative humidity are 279.3 K and 61.6 % in comparison to the observed values of 280.3 K and The observed and modeled distributions of AOD and PM are compared in Fig. 6. The AOD observations based on the LE on average are ∼ 50 % larger than those from the SP. Modeled AOD on average are ∼ 20-30 % higher when compared to SP at the Bishkek and LST sites and ∼ 1 % lower when compared to the LE values. The variability in the predictions is slightly under-estimated. PM 2.5 is over predicted (∼ 50 %) and the 10 spread is accurately captured, while PM 10 is over predicted by ∼ 70 %. This leads to an underestimation of the PM 2.5 /PM 10 (0.4 predicted vs. 0.5 observed) and also helps account for the overestimation in modeled AOD (by ∼ 20-30 %).
Chemical analysis of the filter and soil samples in the CA dust regions have been used to estimate the dust contribution to measured PM at the two sites and to help 15 identify source regions of importance (Park et al., 2014). The emission regions within CA, including around the Aral Sea, and western China were identified as the most important dust sources, which is consistent with the regions identified in the simulations. Dust was estimated to comprise between 5-40 % of PM 2.5 mass at the LST site and to vary by season (minimum values in winter). The observation-based estimates of dust 20 % contribution suggest that modeled dust is over predicted by ∼ 2 times. Thus it appears that dust is a main reason for the over prediction of PM 2.5 and PM 10 , and that dust emission models need to be refined for CA applications.
The overestimation in PM mass at the surface could also be impacted by errors in the modeled PBL heights. The PBL height as determined by the Lidar aerosol profiles 25 varies seasonally and are highest in the summer (from 2-4 km a.g.l.) and lowest in the winter (November-February, 0.5-1.5 km a.g.l.) (Fig. S5). The predicted PBL heights show a similar seasonal cycle with a tendency to under-predict the heights in all seasons as indicated by the comparison of the distributions of the observed and predicted 11357 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | values (Fig. 5), and this occurs in all seasons (Fig. S5). The lower PBL height in the model contributes a systematic high bias in surface concentrations driven by near surface emissions.
Further statistical details of the model-observation comparison can be found in the Supplement (Tables S2 and S3). The seasonal variability in the observations is dis-5 cussed in further detail later (Sect. 3.4).

Source contributions to PM 2.5
Model simulations were performed to identify the component, source region and emission sector contributions to PM 2.5 mass. Period means for the spatial average over the entire CA region (see Fig. 1) and for the grid cells for the Bishkek and LST observation 10 locations are presented in Fig. 7, and their comparison provides insights into the spatial variability of PM and its sources within CA, and how representative the observation sites are at characterizing CA PM. The component contribution to AOD at the sites and for the CA average are similar, with the major contributions coming from fine dust, sulfate, and OC. Spatial maps of mean percent contributions of the various components 15 (i.e., BC, OC, sulfate, OPM, dust and sea salt) to AOD and PM 2.5 mass are presented in Figs. S6 and S7, respectively. Coarse particles contribute ∼ 10 % to mean AOD. Dust accounts for > 60 % of the calculated PM 2.5 mass at the observation sites and for the CA region. The dust source regions (see Fig. 1) contributing to PM 2.5 vary within CA. Dust from the CA source regions has the largest influence on the region mean 20 dust-PM 2.5 mass. At the LST site, which is located in the far east of CA, western China dust sources have their largest influence (∼ 40 %). African and Middle East source regions have their largest influence on the Bishkek site (20 and 15 %, respectively), and collectively contribute ∼ 25 % to regional CA dust PM 2.5 .
The source region contributions to the non-dust PM 2.5 are very similar for the Bishkek 25 and LST sites, with CA sources making the largest contribution (∼ 50 %) followed by Europe (∼ 20 %), the Middle East (∼ 15 %), and biomass burning (∼ 15 % from all sources). For the entire CA region the European source contribution is as large as 11358 3 to 7 % to PM 2.5 in CA. Of the biomass burning contribution to PM 2.5 , the Siberian and European fires (see Fig. 1 for fire regions) contribute 63 and 25 %, respectively, with contributions from South/Southeast Asia and North America fires each contributing ∼ 5 %. The power and industrial sectors are identified by the simulations as the largest contributors (∼ 40 % each) to non-dust PM 2.5 mass in CA. There is also a clear seasonality in the surface meteorology in the region as shown by the time series in surface temperature and relative humidity at the two sites (Fig. 10). There are distinct temperature minima in the winter and relative humidity minima in the summer. However there is not a clear seasonality in wind speed and direction, and the 25 winds are generally from the south and less than 4 m s −1 throughout the year at the LST site (not shown).
The source region and component contributions exhibit seasonal variability as shown by the modeled contributions to PM 2.5 mass in Fig. 11. Dust is found to be the main driver of the seasonal cycle of PM 2.5 . The dust contribution to PM 2.5 is peak in spring and minimum in winter (< 20 %). During this time period the transport of air masses to the sites are from the west and the southwest. When the transport is from the east 5 then dust sources from western China can impact the stations. This transport pattern occurs episodically throughout the year, with contributions from western China sources as large as 20 to 50 %. The dust seasonal cycle is in turn is influenced by the seasonal variations in meteorology that drives the dust emissions and transport. The seasonal changes in the dust source regions can be seen in the seasonal spatial maps of AOD 10 ( Fig. S8). Throughout the domain, AOD in the dust regions are highest in March-October and lowest in winter (Fig. S8) as the nearby dessert regions are snow covered.
Biomass burning also adds to the seasonal cycle, and its contribution is minimum in the winter. South Asia sources can impact the sites in the winter time. The periods when North America sources impact the site are associated with strong transport 15 events across the Atlantic and subsequent subsidence towards the surface associated with high pressure systems as they move towards CA. The transport pathways are discussed in more detail in Sect. 3.7.

Source contributions to BC
Because of its dual role as an air pollutant and as a climate warming agent there is spe-20 cial interest in understanding the regional and sector contributions to BC (Ramanathan and Carmichael et al., 2008). BC comprises on average only about 1-2 % of PM 2.5 mass in CA. The period mean predicted BC surface concentrations are ∼ 0.1 µg m −3 at the two observation sites and 0.15 µg m −3 for the CA regional average. As shown in Fig. 7 contrast to the sector contributions to non-dust PM 2.5 mass, where power and industry are the most important sectors. On average biomass burning contributes ∼ 10 % to BC mass, with Siberian and European fires accounting for 61 % and 33 %, respectively. The source contributions to OC are shown in Fig. S9. There is also large seasonal variability in BC concentrations and source sec-5 tor/region contributions (Fig. 12). BC surface concentrations show the highest values in fall/winter (as do the observations), when there is maximum contribution from the residential sector, reflecting the wide-spread use of biofuels and coal for heating in the region. The source region contributions vary by season, with maximum contributions from Europe and China. South Asia sources contribute in the winter. Biomass burning 10 also is an important source of BC and plays an important role in influencing daily and seasonal variability in BC concentrations. Predicted BC captures the seasonality and the magnitude of the spring and summer values as observed, but concentrations are biased low in the fall/winter. Median BC concentrations (and variability) are underestimated by a factor of 2 at both observation 15 sites (Fig. 6 and Tables S2 and S3). The high wind speed bias in winter (∼ factor of 2), should result in too rapid dispersion and could contribute to the negative bias, but the negative bias in the PBL heights should lead to higher predicted concentrations. Thus this negative bias is likely related to emissions (an indication of an underestimation of the heating fuel use).

20
The OC concentrations follow a similar seasonal cycle as BC and are also under predicted ( Fig. 6 and Tables S2 and S3). Furthermore the OC/BC ratio is under predicted by a factor of ∼ 3 (Fig. 6). The observed OC/BC ratio follows a seasonal cycle with values > 15 in summer and ∼ 5 in September through April. Part of this under prediction in OC and the OC/BC ratio is due to the fact that SOA is not estimated 25 in the model. However a source contribution of OC using the filter data and chemical mass balance (CMB) approach found that SOA sources were very low in winter and only ∼ 20 % in summer (Miller-Schulze et al., 2011). Thus SOA cannot account for the model under prediction of winter values. There appears to be an underestimation of Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | regional OC primary emissions. SOA can however help account for the large values of OC/BC observed in the summer and not predicted. Biomass burning emissions cannot account for the underestimation in winter BC and OC. The largest impact of fires at the observation sites is in the late summer, when the fires are concentrated in western Russia and the wind direction is such that the smoke 5 is transported into CA. Fires from South Asia can impact the sites associated with the fires and high pollution levels in northern India and with winds from the south, which can occur in late fall, but not frequently.
The fact that BC and OC are systematically under predicted in the winter suggests that local/regional emissions during the heating season may be underestimated. This 10 is supported by the results of the CMB analysis of OC discussed above that found the contribution from biofuel combustion increased 2-3 times in the fall and winter periods. The uncertainty in emissions can also be partly caused by the lack of seasonal emissions over this region as described earlier in Sect. 2.

Vertical distributions
15 Figure 13 shows the predicted weekly averaged vertical distributions of PM 2.5 , dust, and BC for the entire simulation period at the LST site. These plots show more clearly that much of the variability in the PM loadings is associated with dust and biomass burning episodes (as represented by the enhancements in BC). Typically the high PM episodes show elevated PM mass that extend from the surface to 2 to 4 km. The vertical 20 extents show a seasonality associated with seasonal variations in the PBL heights. These vertical distributions indicate that much of the transport of aerosols in CA occurs via low altitude pathways. In some cases there are large amounts of dust and biomass burning aerosol in the 3-6 km altitude range that are decoupled from the surface (e.g., dust in early May 2009), reflecting that some aerosols are lifted out of the boundary 25 layer and are transported at high altitude over CA, enhancing AOD but not contributing to ground-level mass concentrations at the observation sites. These vertical features are confirmed by the aerosol extinction profiles observed at the LST site as discussed 11362 Introduction  Chen et al. (2012b). The variation in weekly averaged AOD can be significant (Fig. 13 bottom panel) and is driven by variations in dust and biomass burning emissions.

Transport pathways
The three dimensional ten day air mass trajectories (described in Sect. 2.2) were utilized to further understand the transport pathways of air masses entering into and ex-5 iting out of the CA region and its subsequent impact of source regions on the aerosol distributions at the CA sites.

Transport into CA
The air mass transport into CA is discussed through back trajectories associated with the five events labeled on Fig. 13. These five events represent transport episodes with elevated surface PM 2.5 (averaged over the three hour time window consistent with trajectory time step) with varying contributions from biomass burning, anthropogenic pollution, and dust sources. In each trajectory figure (Figs. 14 and 15), the regions with active dust (blue diamond hatches) and biomass burning emissions (green square hatches) for the event time period and prior ten days are identified and MODIS AOD ized by high levels of BC without dust. During this episode the transport to the site was under the influence of a high pressure system located to the northwest and air masses were transported over the active fire region in western Russia. Figure 15 shows winter and spring events. The November episode (event 3) is a period with elevated BC and PM 2.5 from pollution sources from South Asia (including 5 some fires) and western China and low fire and dust emission activity. The January episode (event 4) is a period of elevated BC with air masses coming from Europe, indicating the influence of anthropogenic pollution coming from this industrialized region, and from CA sources. Dust emissions from CA and Africa were low during this period. The final illustrative episode is for April 2009 (event 5), a period with both elevated lev-10 els of dust from western China, CA and Africa and BC from both fire and anthropogenic pollution from Europe, CA and Russia sources.
These examples provide insights into the source region contributions to PM mass in CA as presented in Figs. 11 and 12). CA is an ideal location to observe a variety of source regions as it is at the crossroad of transport patterns with air masses impacted 15 from dust, anthropogenic activity and biomass burning from different geographical regions.

Long range transport of CA sources
The transport pathways out of CA were also evaluated by calculating forward trajectories from the observation sites. Selected forward trajectories initialized at or below 20 1 km are used to represent the transport of boundary layer PM from CA and these are shown for summer, winter and spring periods in Fig. 16. In these plots the MODIS AOD, dust, and fire emissions plotted for each event represent values averaged over the subsequent ten days and trajectories were stopped if they impacted the surface. During the summer, outflow from CA is towards the north in association with the sum-25 mer monsoon system. Figure 16a shows the subset of forward trajectories that reside for at least 3 days over the region 48-65 • N during June 2008. Trajectories typically pass over Russia and reach into the Arctic and also can be caught in westerly storm 11364 Plotted are the forward trajectories that stay within the 30-50 • N region for at least 3 days during the month of April 2009. During the spring transport from CA is dominated by strong westerly flows and air masses are transported over China, Korea, and Japan and then across the Pacific, reaching North America in 7-10 days. PM arising from dust and anthropogenic emissions from CA impact the entire North- There are episodic and seasonal components to the intercontinental transport as shown in the time series of the vertical profiles of PM 2.5 at Mt. Bachelor,Oregon (43.97 • N, 121.69 • W, 2700 m a.g.l.) (Fig. 18). The bulk of the CA particle transport takes place in the free troposphere and impacts surface concentrations in the US as 20 the boundary layer grows and entrains "plumes" aloft. This occurs most frequently in spring, summer and fall. The episodic contributions of CA sources to surface concentrations can exceed 1.5 µg m −3 . In the fall, there is also strong transport of dust from CA across Europe and out into the Atlantic.

Future scenarios 25
How might PM levels change in CA over the next few decades? To address this question, simulations were repeated for various emission scenarios developed and used in the WMO/UNEP assessment on short-lived climate pollutants (Shindell et al., 2012) 11365 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | as described in Sect. 2. Dust and biomass burning emissions and meteorology were the same as those used in the 2008/2009 simulations. The period mean changes in surface BC and PM 2.5 concentrations in 2030 for the reference scenario are shown in Fig. 19a and d, respectively. This scenario reflects all present agreed policies affecting emissions and assumes that they are fully implemented. Under this scenario PM 2.5 5 increases significantly in South Asia and western China (> 50 %) and in parts of CA, including the area where the sampling sites are located. PM 2.5 decreases in Western Europe and Eastern China (< 10 %). BC surface concentrations show a similar pattern to PM 2.5 , although covering larger portions of CA with relatively larger increases in BC than in PM 2.5 . These results suggest that health impacts and climate warming due 10 to BC and PM 2.5 may increase in coming decades unless additional emission control measures are implemented. Results for two other scenarios are also presented in Fig. 19. One scenario specifically targets BC emission reductions in recognition that BC is also a major contributor to atmospheric warming (Ramanathan and Carmichael, 2008). These additional mea- 15 sures significantly reduce 2030 BC concentrations by greater than 35 % throughout most of the domain, with only a few regions (e.g., Myanmar and eastern Afghanistan) showing increases in BC relative to 2005 levels. This scenario assumes that all BC emission reduction measures are perfectly implemented and 100 % effective. BC measures also impact emissions of co-emitted pollutants (e.g., OC and SO 2 ). PM 2.5 con-20 centrations under this scenario (Fig. 19e) are reduced, but by much smaller amounts, and concentrations still increase relative to 2005 over large regions of South Asia and western China, and parts of CA. These results suggest that health impacts in these regions may increase due to the PM 2.5 increases whereas positive radiative forcing and health effects due to BC may decrease. When the BC measures are used along with 25 greenhouse gas measures aimed at keeping CO 2 levels below 450 ppm, the PM 2.5 levels in South Asia are lower than 2005 levels (Fig. 19f), with few exceptions (one being Myanmar region). This is due to the large decreases in SO 2 and NO x emissions under Introduction  The seasonal cycles and source sector and source region contributions to PM in CA were analyzed using the STEM chemical transport model. Dust was the largest component of the PM 2.5 and PM 10 mass in the region in all seasons except winter, whereas sulfate was the largest anthropogenic component of the PM 2.5 mass. Dust was also 15 found to be the major driver of the seasonal cycles of AOD and PM concentrations. On an annual basis the power and industrial sectors were the most important contributors to PM 2.5 , while residential and transportation were the most important sectors for BC. Open biomass burning within and outside the region also contributed to elevated PM and BC concentrations and to the temporal variability.

20
The model simulations showed a systematic over prediction of PM mass. This is most likely due in large part to the over prediction in dust. Carbonaceous PM was underpredicted and it is speculated that the winter emissions associated with residential heating may be underestimated in the current emissions inventory. The predicted wind speeds were biased high (by ∼ 30 %) and the direction had a southwest bias. The high bias 25 in wind speeds may also contribute to the over-prediction in PM 10 , as dust emissions depend strongly on wind speed. Efforts to improve the dust emissions and to improve the wind speed and direction predictions using a finer model resolution are planned. Additional efforts are needed to improve the anthropogenic emissions estimates for CA.
Currently there are few measurements in CA that can be used to quantify the intercontinental transport of pollution from Europe to Asia. The analysis of the transport 5 pathways and variations in PM mass and composition observed at the two sites in CA demonstrate that this region is strategically located to characterize regional and intercontinental transport of pollutants. Aerosols at these sites were shown to reflect dust, biomass burning, and anthropogenic sources from South, East, and CA, Europe, and Russia depending on the time of year. For example, during the spring fine particles 10 from Europe and Africa were transported to CA, on to eastern Asia, and then across the Pacific to North America.
Observations of PM and its composition in this region are of growing importance as it is estimated that PM 2.5 levels are likely to increase significantly in Central and South Asia and western China over the next few decades. Simulations for a reference 15 2030 emission scenario showed that BC concentrations had a larger relative increase than PM 2.5 concentrations. This suggests that health impacts and climate warming associated with these pollutants may increase over the next decades unless additional control measures are implemented. Continued pollutant observations in CA will help to characterize the changes that are rapidly taking place in the region.